This week I earned three OpenAI certifications: AI Foundations, Applied AI Foundations and Agents and Workflows. I’ve been using AI every day for months, building marketing tools and automations with Claude Code. But using a tool and understanding it are two different things, and I wanted both.
The courses filled in the gaps. They gave me clearer language for what AI is actually doing, better habits for using it well, and a framework for the part that often gets skipped: the responsibility that comes with it.
This post is what I took away, written for marketers and small business owners who use AI (or are about to) and want to get it right.
The foundations: what AI actually is
Tools like ChatGPT and Claude are built on large language models. In simple terms, they’ve learned patterns from enormous amounts of text, and they generate a response by predicting what should come next. That’s what makes them so fluent, and it’s also why they need a human in charge.
Three things changed how I think about it:
- It isn’t a search engine. An AI model doesn’t look up facts by default. It produces the most likely answer, which is usually right but sometimes confidently wrong. These wrong answers are called hallucinations, and they read just as convincingly as correct ones.
- It only knows what it’s given. A model has a training cut-off, and it knows nothing about your business, your client or your brand unless you tell it. The quality of the output depends heavily on the context you provide.
- It’s a collaborator, not an oracle. AI is brilliant at drafting, summarising, restructuring and brainstorming. It’s less reliable at final judgement calls, exact figures and anything that needs to be true.
Once you understand that, the right way to use it becomes obvious: give it good context, check what matters, and keep the final decision with you. When I built my marketing dashboard, the biggest jump in quality didn’t come from clever prompts. It came from giving the tools our real material: our CI rules, logos, case study template and past content.
Applied AI: using it well every day
The difference between a disappointing result and a useful one is rarely the tool. It’s how you brief it. I now treat every prompt like a brief I’d give a new team member:
- Context: who it’s for, what the business does, and any background that matters.
- Task: exactly what you want, in one clear sentence.
- Constraints: tone, length, format, words to avoid, and anything that must be included.
- Examples: a sample of what good looks like. One example teaches more than a paragraph of description.
- Iteration: treat the first answer as a draft. Ask for changes, push back and refine.
For marketing, the biggest win is consistency. Instead of starting from scratch each time, I save the context once, such as our brand voice, audience and formatting rules, and reuse it. That’s exactly how the content generators in my dashboard work: the brief for a blog post, social post, emailer or case study is saved once, so anyone on the team gets an on-brand first draft in minutes and spends their time on the thinking rather than the rewriting.
Agents and workflows: from chatting to doing
Most people use AI as a chat window: ask a question, get an answer. The next step is using it to do things.
- A workflow is a series of steps you’d normally do by hand, connected so they run in order. For example: take a webinar transcript, pull out the key points, draft a blog post, then create three social posts from it.
- An agent goes further. You give it a goal and access to tools, and it decides which steps to take, such as searching, reading files, writing and checking its own work.
This is where I’ve seen the biggest time savings in my own work.
The key lesson: automate the repeatable, keep humans on the judgement. Map the process first, decide where a person needs to check or approve, and start small. A workflow that saves an hour every week is worth more than an ambitious agent nobody trusts.
Putting it into practice: my marketing dashboard
Before the courses, I’d already built a marketing dashboard with Claude Code for my team. Looking back, it’s a working example of almost everything they cover. I built it for three reasons: to cut the time spent on repeat tasks, to keep every marketing resource in one place, and to track tasks and deadlines.
- One source of truth. Our CI rules, logos, templates and case studies live in one place, so the team, and the AI tools, always work from the latest version.
- Saved briefs. Generators for case studies, blog posts, social posts, emailers and transcripts turn a short brief into a first draft that already follows our brand.
- Workflows that save real time. A customer report that used to take over half an hour now exports in about five minutes. A case study can go from brief to draft to published through one connected process.
- Structure that keeps things moving. Weekly cadence tasks and a client view keep the team and our clients on track without anyone having to chase.
- Built with the team, not for them. I mapped how we actually work first, shared an early version, and let the team’s feedback shape the next one.
The biggest lesson: the tool is only as good as the process behind it. Knowing your workflow is the real skill. AI just makes it faster.
Using AI responsibly
The more useful AI becomes, the more it matters how we use it. These are the responsibilities I now build into every project:
- Check before you publish. You’re accountable for what goes out under your name, not the tool. Verify facts, figures, names, dates and claims, especially anything a client or customer will rely on.
- Protect people’s data. Don’t paste client information, customer lists or anything confidential into a tool unless you know how it’s stored and your agreement allows it. In South Africa, POPIA applies to personal information whether you process it yourself or through an AI tool.
- Be open about how you use it. Clients deserve to know when AI is part of the process. Being upfront builds trust, and it’s increasingly expected.
- Watch for bias. Models learn from human-made data, including its blind spots. Review content for stereotypes, one-sided views or language that excludes people.
- Respect other people’s work. Don’t ask AI to copy a competitor’s content, imitate a specific artist or reproduce copyrighted material. Use it to create something original.
- Keep a human in the loop. Automate the steps, but keep a person responsible for decisions, approvals and anything sent to clients.
In my own dashboard, that means every generator produces a draft, not a final product. A person reviews everything before it’s published or sent to a client, and our CI and brand voice rules are built in so the tools can’t drift off-brand.
None of this slows you down much. It’s the difference between using AI quickly and using it well.
What’s next
These certifications didn’t change what I do with AI. They changed how deliberately I do it. I understand the tools better, I brief them better, and I build responsibility into every workflow from the start.
If you’re a small business or marketing team wondering where to begin, start with one repetitive task, give AI clear context, and check the result like you would a colleague’s work. That small habit is the foundation for everything else.
If you’d like help bringing AI into your marketing the right way, get in touch.
